Adaptive Mutations in Bacteria: High Rate and Small Effects

See allHide authors and affiliations

Science  10 Aug 2007:
Vol. 317, Issue 5839, pp. 813-815
DOI: 10.1126/science.1142284


Evolution by natural selection is driven by the continuous generation of adaptive mutations. We measured the genomic mutation rate that generates beneficial mutations and their effects on fitness in Escherichia coli under conditions in which the effect of competition between lineages carrying different beneficial mutations is minimized. We found a rate on the order of 10–5 per genome per generation, which is 1000 times as high as previous estimates, and a mean selective advantage of 1%. Such a high rate of adaptive evolution has implications for the evolution of antibiotic resistance and pathogenicity.

The rate at which new mutations arise in natural populations and their fitness effects are of key importance in evolutionary genetics. Classical mutation accumulation experiments have indisputably shown that among the spontaneous mutations that affect fitness, those that cause deleterious effects are far more common than those that cause increases in fitness. Whereas there are currently several direct and indirect estimates of the deleterious mutation rate in different organisms, data are lacking for beneficial mutations (1). The latter are of particular interest because they constitute the driving force of adaptation and survival of populations in new environments.

Several theoretical studies have made some general predictions about the long-term process of adaptation toward an optimum (2, 3). One prediction suggests that the effects of beneficial (advantageous) mutations (sa) are exponentially distributed, in that many have very small effects and those that cause strong increments in fitness are rare (3). These are plausible predictions given that organisms are in general well adapted to their environments, so only small and rare changes lead to fitness increases (411).

The true distribution of newly arising beneficial mutations in an organism in a given environment is difficult to estimate because the probability of fixation of a beneficial mutation that increases fitness by sa is only 2sa, which means that mutations of small effect are not likely to increase in frequency. This implies that the distribution of mutations that escape stochastic loss (become fixed or reach high enough frequencies to be observed) is truncated for small values (12, 13). In addition, clonal interference occurs in large populations with a high beneficial mutation rate (Ua) and no recombination and will slow adaptation (compared to sexual populations of the same size) (14). Namely, if multiple beneficial mutations appear in different lineages, they compete with each other for fixation. This translates into an adaptation rate less than that predicted by the mutation rate and population size, and into the fixation only of mutations of large effect (15). Recently, there has been a considerable effort to predict the rate and distribution of beneficial mutations and the effect of clonal interference on the adaptation rate (16, 17).

Current estimates for Ua fall around 10–9 to 10–8 for RNA viruses and Escherichia coli (4, 5, 16). A similar beneficial mutation rate was estimated for Pseudomonas fluorescens under adaptation to stressful conditions (9). A caveat for all of these estimates is that they were obtained from populations with very large effective population size (Ne) and followed adaptation to a new environment under conditions in which clonal interference had a strong effect. This led to downward biased estimates of Ua. Here, we provide estimates for the genomic mutation rate for beneficial mutations in E. coli that are less biased by clonal interference.

In this work, we used populations with an intermediate effective population size—big enough that genetic drift is unlikely to drive slightly deleterious mutations to a high frequency but small enough to minimize the effects of clonal interference between beneficial mutations. To estimate the beneficial mutation rate and the distribution of fitness effects of single mutations, we used a microsatellite marker system pioneered by Imhof and Schlotterer (4). Mutations at a microsatellite locus coded by a nonconjugative plasmid can generate neutral allelic diversity in a very short time (4, 18), and selective sweeps, occurring in the bacterial genome, can be identified by following the rapid increase in the frequency of the linked microsatellite allele (4).

We allowed populations of E. coli to adapt to a given laboratory environment for 1000 generations and followed the allelic distribution of the microsatellite at periodic intervals. From this distribution, the number of mutations that escaped stochastic loss during this period was inferred for populations with a small effective size (Ne = 2 × 104) and for populations with a very large effective size (Ne = 107). The latter allowed us to compare our estimates with those previously published (4, 16, 19).

The beneficial mutations that escape stochastic loss are expected to follow a gamma distribution with shape parameter 2 and with a mean equal to twice that of the distribution of the spontaneously arising mutations (16). Figure 1 shows the observed distributions of effects of favorable mutations segregating in the populations. In the populations with the smaller effective size (Ne = 2 × 104), the mean value of the selection coefficient [E(sa)] measured was 0.013, which is slightly smaller, although close to previous estimates (4, 16). In these populations (Fig. 1A) we find that a gamma distribution with such parameterization provides a good fit to the data (Kolmogorov-Smirnov: not significant, P = 0.6). In Fig. 1B, we show the distribution of selective effects measured in the populations with larger effectivesize(Ne = 107). As expected, in these populations, the effect of clonal interference was clearly observed in the distributions of microsatellite allelic variation [for an example, see fig. S2 (19)]. As predicted theoretically (15), the effect of interference between clones carrying different beneficial mutations is reflected in an increased value of the mean selective effect of mutations segregating in the population [E(sa) = 0.023, as shown in Fig. 1B]. This is because many newly arising beneficial mutations of small effect are lost in competition with mutations of larger effect.

Fig. 1.

Distribution of fitness effects of beneficial mutations that escaped stochastic loss, measured in populations of Ne = 2 × 104 (A) and Ne = 107 (B). The gray bars show the distribution of effects of beneficial mutations inferred in the experimental populations and the white bars correspond to a gamma distribution with shape 2 and scale parameters 158 (A) and 85 (B). Both distributions are supported by the data [Kolmogorov-Smirnov: P = 0.6 in (A) and P = 0.5 in (B); not significant].

To measure the rate of spontaneous beneficial mutations, we quantified the total number of mutations that escaped stochastic loss in all the populations with the same effective size during the course of the experiment. We observed 75 such events in the populations with Ne = 2 × 104 and 87 in the populations with Ne = 107. Assuming that the effect of clonal interference is negligible, in the populations with larger Ne we would infer a mutation rate of 2 × 10–8 beneficial mutations per genome, per generation (20, 21). This value is close to those previously measured for this species with the use of populations with similar effective sizes (4, 16). However, with such a large Ne, the effect of clonal interference is very important and leads to an extreme underestimation of the true value of Ua. Indeed, in the populations of smaller effective size, our estimate of the mutation rate was 1000 times as high: Ua = 2 × 10–5 beneficial mutations per genome, per generation (20, 21). Given that clonal interference is much weaker in these populations, we take this value to be a much more accurate measure of the real Ua.

To complement these results, we measured the mean fitness of each evolved population relative to the ancestral one. Mean fitness of an evolved population was assessed by its competitive ability against a reference strain (19). As expected in view of the results obtained above, there was an overall increase in fitness in all populations after 1000 generations of adaptation. In the populations with the smaller Ne, this increase was about 17%, which, as expected, was smaller than the one observed for the populations with a larger Ne (overall mean increase in fitness of 40%). We then asked if our estimates of Ua could explain such increments in fitness (22). To do this, we compared the results of Monte-Carlo simulations of adaptive evolution, assuming several different values of Ua and E(sa) with those obtained in the experiments (19). In all the simulations, we assumed that the distribution of incoming beneficial mutations is exponential and that mutations interact in a multiplicative way (2, 23). Different combinations of Ua and E(sa) were consistent with the fitness increase in populations of a given effective size, but the set of parameters that more closely matched the combined data of both population sizes was Ua between 10–5 and 10–4 and E(sa) between 1 to 2% (Fig. 2). These parameters agree with the estimates obtained from the microsatellite allelic distribution (small effective size populations in Fig. 1, in which the measured mutation rate was 2 × 10–5). It is also clear that a mutation rate of about 10–9 or 10–8 [as inferred in other experiments (4, 16)] cannot explain the fitness increases observed.

Fig. 2.

Mutation rates (Ua) and mean effect of beneficial mutations [E(sa)], used as parameters in Monte Carlo simulations (19), which produced mean fitness increases consistent with those observed in the evolved populations (difference was not significant; Student's t test P > 0.05). The circles show the parameter values consistent with the mean fitness observed in the populations of Ne = 2 × 104 and the triangles in the populations of Ne = 107.

Our results show that the mutation rate to new beneficial alleles is 1000 times as high as previously inferred in the same bacterial species (4, 16). The difference in results can be explained by the differences in the effective population size analyzed. If only very large effective sizes are analyzed, and the effect of clonal interference is not accounted for, then our estimates for Ua and E(sa) for the populations with Ne = 107 are similar to those previously obtained (Fig. 1B). However, if these estimates were close to the true values, then we would not expect to see the sweeps of beneficial mutations in the populations with lower Ne that we observed (Fig. 1A). Hence, neglecting the effect of clonal interference underestimates the value of Ua. In addition, we showed that clonal interference changes the distribution of segregating mutations: When comparing the distribution of beneficial mutations for the populations with high Ne (strong effect of clonal interference) with that for populations with low Ne (Fig. 1, A and B), a significant difference was observed (Kolmogorov-Smirnov: P = 0.001). As predicted theoretically (15, 24), we observed a distribution with a higher mean selection coefficient when the effect of clonal interference was stronger. In the limiting case where the supply of new beneficial mutations per generation (NeUa) is very high, the speed of adaptation will no longer depend on NeUa but on the mutations of largest effect available, because these are the only mutations that will fix. This might help explain why similar beneficial mutation rates are estimated in very diverse organisms under very diverse environments. These estimates are obtained in populations with very large effective sizes (4, 5, 9, 16), which are likely to produce strong underestimations of Ua.

It is plausible that, in natural habitats, population sizes will be large. If the effective size of a bacterial species is much higher than 104 (25), then our results imply that clonal interference plays a major role in limiting the adaptation of these asexual organisms. As such, if there is a chance for recombination, clonal interference will be much lower and organisms will adapt faster. This has been predicted theoretically (14), although the empirical evidence is still very preliminary (26, 27). Given our results, we anticipate that clonal interference is important in maintaining sexual reproduction in eukaryotes. Notably, mutation accumulation experiments in Saccharomyces cerevisiae and Arabidopsis thaliana have detected a significantly large number of mutants with increased fitness (28, 29).

Given the estimates for the overall mutation rate in E. coli (30) and its genomic deleterious mutation rate (1), our estimate of Ua implies that 1 in 150 newly arising mutations is beneficial and that 1 in 10 fitness-affecting mutations increases the fitness of the individual carrying it. Hence, an enterobacterium has an enormous potential for adaptation and may help explain how antibiotic resistance and virulence evolve so quickly.

Supporting Online Material

Materials and Methods

SOM Text

Figs. S1 to S4

Table S1


References and Notes

Stay Connected to Science

Navigate This Article